
Radu Stefan Popa
I am a final-year Computer Science student at the University of Bucharest. I work on multimodal models and language models. With Radu Tudor Ionescu I work on vision-language models across languages, and with Sergiu Nisioi on small language models for Romanian. Since I started research in May 2026, I have worked on language models, image captioning and image generation, and each time I started from a problem that the group needed solved.
Alongside my studies I work full-time as a software engineer at Liamis & HEB International and teach weekly algorithms classes at my faculty.
Papers
Ongoing
PaMuk-BenchA parallel benchmark for vision-language tasks in seven languages, with Radu Tudor Ionescu and a team of 13 researchers. I led the captioning study and built the method that taught three image generators six more languages, by distilling each model's own English behavior into its text encoder.